On the removal of random variables in data sets of meteorological observations

For some purposes it is not necessary to use a full data set of strongly spatially correlated parameters. In some situations, too much information may even be unwanted. We propose a procedure for finding a non-redundant choice of synoptic stations that is sufficient to capture the relevant physical patterns based on rule N. The technique makes use of principle component analysis and cluster analysis. The above-mentioned procedure can be applied to generic data sets. As an illustration, we apply it here to the strongly correlated two-meter temperature observed in the Belgian synoptic network during the winter period. We find that about three or four stations from this network are necessary. The method also suggests an intelligent choice of stations that are most suitable to be used.